sfx2000
Part of the Furniture
We'd need Broadcom/QCA to either provide some form of API to allow a packet scheduler (like Cake) get access to flows, or for them to start integrating support for SQM into their NAT acceleration engine.
Outside of just Broadcom/QCA - challenge with flow offload and QoS is that flow offload promises every pack will be routed, where as QoS can, and often does, drop packets to keep each flow managed to the desired characteristics...
Cake SQM will definitely start tossing packets out the back if a higher priority flow needs bandwidth.
These x86 boxes have to do fast path and QoS management all in software - there, clock speed definitely matters more than number of cores